Support Export in Web

[rebornix]

This is part of #9574

Currently the Export actions are powered by nbconvert. We will run nbconvert in a separate process and save the converted content in a temporary file on disk. This is not feasible on Web as Jupyter runs in web worker and there is no jupyter/nbconvert in the same environment. To archive the same thing on Web, we would need to ask the remote Jupyter server/service to handle the nbconvert request for us.

There are two ways to run nbconvert on the remote Jupyter server:

• The first one is directly calling the nbconvert web service, we can construct an url as ${baseUrl}/nbconvert/${type}/${filePath}?token=${token} and fetch this resource in the web extension. However it requires the file to be synced to the remote Jupyter server
• The second one is calling nbconvert through shell commands
  • We can either run it against files, i.e., !jupyter nbconvert test.ipynb --to python --stdout, but it has the same problem as the first approach
  • Or we can pass in the content through stdin: !echo '{"cells": [], "metadata": {}, "nbformat": 4, "nbformat_minor": 2 }' | jupyter nbconvert --nbformat 4 --stdin --to python --stdout. This is the only approach that doesn't require file to be synced but if the file is large, it will affect the throughput.
  • The catch with running shell commands is it requires us to know the path of the active jupyter executable as it might exist in PATH and it needs to be platform specific. We might need to run {sys.executable} -m jupyter nbconvert

[rchiodo]

We've used the {sys.executable} trick in the past to run python on the jupyter server.

Another idea? Zip the file on the local machine, send zip with jupyter file transfer API, reopen zip in temp folder and run nbconvert on it?

The stdin way sounds plausible but I'm worried about gb size notebooks (that zip up to mbs)